diff --git a/target-mips/cpu.h b/target-mips/cpu.h
index d127d00e8f..e0f03eda73 100644
--- a/target-mips/cpu.h
+++ b/target-mips/cpu.h
@@ -81,9 +81,9 @@ struct CPUMIPSState {
 #define FCR0_REV 0
     /* fcsr */
     uint32_t fcr31;
-#define SET_FP_COND(num,env)     do { (env->fcr31) |= ((num) ? (1 << ((num) + 24)) : (1 << ((num) + 23))); } while(0)
-#define CLEAR_FP_COND(num,env)   do { (env->fcr31) &= ~((num) ? (1 << ((num) + 24)) : (1 << ((num) + 23))); } while(0)
-#define IS_FP_COND_SET(num,env)  (((env->fcr31) & ((num) ? (1 << ((num) + 24)) : (1 << ((num) + 23)))) != 0)
+#define SET_FP_COND(num,env)     do { ((env)->fcr31) |= ((num) ? (1 << ((num) + 24)) : (1 << 23)); } while(0)
+#define CLEAR_FP_COND(num,env)   do { ((env)->fcr31) &= ~((num) ? (1 << ((num) + 24)) : (1 << 23)); } while(0)
+#define GET_FP_COND(env)         ((((env)->fcr31 >> 24) & 0xfe) | (((env)->fcr31 >> 23) & 0x1))
 #define GET_FP_CAUSE(reg)        (((reg) >> 12) & 0x3f)
 #define GET_FP_ENABLE(reg)       (((reg) >>  7) & 0x1f)
 #define GET_FP_FLAGS(reg)        (((reg) >>  2) & 0x1f)
diff --git a/target-mips/exec.h b/target-mips/exec.h
index e82b71651b..7ebfd78e71 100644
--- a/target-mips/exec.h
+++ b/target-mips/exec.h
@@ -165,4 +165,75 @@ void cpu_mips_update_irq (CPUState *env);
 void cpu_mips_clock_init (CPUState *env);
 void cpu_mips_tlb_flush (CPUState *env, int flush_global);
 
+void do_ctc1 (void);
+void do_float_cvtd_s(void);
+void do_float_cvtd_w(void);
+void do_float_cvtd_l(void);
+void do_float_cvtl_d(void);
+void do_float_cvtl_s(void);
+void do_float_cvtps_pw(void);
+void do_float_cvtpw_ps(void);
+void do_float_cvts_d(void);
+void do_float_cvts_w(void);
+void do_float_cvts_l(void);
+void do_float_cvts_pl(void);
+void do_float_cvts_pu(void);
+void do_float_cvtw_s(void);
+void do_float_cvtw_d(void);
+void do_float_roundl_d(void);
+void do_float_roundl_s(void);
+void do_float_roundw_d(void);
+void do_float_roundw_s(void);
+void do_float_truncl_d(void);
+void do_float_truncl_s(void);
+void do_float_truncw_d(void);
+void do_float_truncw_s(void);
+void do_float_ceill_d(void);
+void do_float_ceill_s(void);
+void do_float_ceilw_d(void);
+void do_float_ceilw_s(void);
+void do_float_floorl_d(void);
+void do_float_floorl_s(void);
+void do_float_floorw_d(void);
+void do_float_floorw_s(void);
+void do_float_add_d(void);
+void do_float_add_s(void);
+void do_float_add_ps(void);
+void do_float_sub_d(void);
+void do_float_sub_s(void);
+void do_float_sub_ps(void);
+void do_float_mul_d(void);
+void do_float_mul_s(void);
+void do_float_mul_ps(void);
+void do_float_div_d(void);
+void do_float_div_s(void);
+void do_float_div_ps(void);
+void do_float_addr_ps(void);
+
+#define CMP_OPS(op)                        \
+void do_cmp_d_ ## op(long cc);             \
+void do_cmpabs_d_ ## op(long cc);          \
+void do_cmp_s_ ## op(long cc);             \
+void do_cmpabs_s_ ## op(long cc);          \
+void do_cmp_ps_ ## op(long cc);            \
+void do_cmpabs_ps_ ## op(long cc);
+
+CMP_OPS(f)
+CMP_OPS(un)
+CMP_OPS(eq)
+CMP_OPS(ueq)
+CMP_OPS(olt)
+CMP_OPS(ult)
+CMP_OPS(ole)
+CMP_OPS(ule)
+CMP_OPS(sf)
+CMP_OPS(ngle)
+CMP_OPS(seq)
+CMP_OPS(ngl)
+CMP_OPS(lt)
+CMP_OPS(nge)
+CMP_OPS(le)
+CMP_OPS(ngt)
+#undef CMP_OPS
+
 #endif /* !defined(__QEMU_MIPS_EXEC_H__) */
diff --git a/target-mips/op.c b/target-mips/op.c
index e119765b3c..5e3c797323 100644
--- a/target-mips/op.c
+++ b/target-mips/op.c
@@ -1609,47 +1609,25 @@ void op_cp1_enabled(void)
     RETURN();
 }
 
-/* convert MIPS rounding mode in FCR31 to IEEE library */
-unsigned int ieee_rm[] = { 
-    float_round_nearest_even,
-    float_round_to_zero,
-    float_round_up,
-    float_round_down
-};
-
-#define RESTORE_ROUNDING_MODE \
-    set_float_rounding_mode(ieee_rm[env->fcr31 & 3], &env->fp_status)
-
-inline char ieee_ex_to_mips(char xcpt)
+/*
+ * Verify if floating point register is valid; an operation is not defined
+ * if bit 0 of any register specification is set and the FR bit in the
+ * Status register equals zero, since the register numbers specify an
+ * even-odd pair of adjacent coprocessor general registers. When the FR bit
+ * in the Status register equals one, both even and odd register numbers
+ * are valid. This limitation exists only for 64 bit wide (d,l,ps) registers.
+ *
+ * Multiple 64 bit wide registers can be checked by calling
+ * gen_op_cp1_registers(freg1 | freg2 | ... | fregN);
+ */
+void op_cp1_registers(void)
 {
-    return (xcpt & float_flag_inexact) >> 5 |
-           (xcpt & float_flag_underflow) >> 3 |
-           (xcpt & float_flag_overflow) >> 1 |
-           (xcpt & float_flag_divbyzero) << 1 |
-           (xcpt & float_flag_invalid) << 4;
+    if (!(env->CP0_Status & (1 << CP0St_FR)) && (PARAM1 & 1)) {
+        CALL_FROM_TB1(do_raise_exception, EXCP_RI);
+    }
+    RETURN();
 }
 
-inline char mips_ex_to_ieee(char xcpt)
-{
-    return (xcpt & FP_INEXACT) << 5 |
-           (xcpt & FP_UNDERFLOW) << 3 |
-           (xcpt & FP_OVERFLOW) << 1 |
-           (xcpt & FP_DIV0) >> 1 |
-           (xcpt & FP_INVALID) >> 4;
-}
-
-inline void update_fcr31(void)
-{
-    int tmp = ieee_ex_to_mips(get_float_exception_flags(&env->fp_status));
-
-    SET_FP_CAUSE(env->fcr31, tmp);
-    if (GET_FP_ENABLE(env->fcr31) & tmp)
-        CALL_FROM_TB1(do_raise_exception, EXCP_FPE);
-    else
-        UPDATE_FP_FLAGS(env->fcr31, tmp);
-}
-
-
 void op_cfc1 (void)
 {
     switch (T1) {
@@ -1675,38 +1653,7 @@ void op_cfc1 (void)
 
 void op_ctc1 (void)
 {
-    switch(T1) {
-    case 25:
-        if (T0 & 0xffffff00)
-            goto leave;
-        env->fcr31 = (env->fcr31 & 0x017fffff) | ((T0 & 0xfe) << 24) |
-                     ((T0 & 0x1) << 23);
-        break;
-    case 26:
-        if (T0 & 0x007c0000)
-            goto leave;
-        env->fcr31 = (env->fcr31 & 0xfffc0f83) | (T0 & 0x0003f07c);
-        break;
-    case 28:
-        if (T0 & 0x007c0000)
-            goto leave;
-        env->fcr31 = (env->fcr31 & 0xfefff07c) | (T0 & 0x00000f83) |
-                     ((T0 & 0x4) << 22);
-        break;
-    case 31:
-        if (T0 & 0x007c0000)
-            goto leave;
-        env->fcr31 = T0;
-        break;
-    default:
-        goto leave;
-    }
-    /* set rounding mode */
-    RESTORE_ROUNDING_MODE;
-    set_float_exception_flags(0, &env->fp_status);
-    if ((GET_FP_ENABLE(env->fcr31) | 0x20) & GET_FP_CAUSE(env->fcr31))
-        CALL_FROM_TB1(do_raise_exception, EXCP_FPE);
- leave:
+    CALL_FROM_TB0(do_ctc1);
     DEBUG_FPU_STATE();
     RETURN();
 }
@@ -1762,45 +1709,31 @@ void op_mthc1 (void)
 
 FLOAT_OP(cvtd, s)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    FDT2 = float32_to_float64(FST0, &env->fp_status);
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvtd_s);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvtd, w)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    FDT2 = int32_to_float64(WT0, &env->fp_status);
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvtd_w);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvtd, l)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    FDT2 = int64_to_float64(DT0, &env->fp_status);
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvtd_l);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvtl, d)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    DT2 = float64_to_int64(FDT0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
+    CALL_FROM_TB0(do_float_cvtl_d);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvtl, s)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    DT2 = float32_to_int64(FST0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
+    CALL_FROM_TB0(do_float_cvtl_s);
     DEBUG_FPU_STATE();
     RETURN();
 }
@@ -1813,81 +1746,55 @@ FLOAT_OP(cvtps, s)
 }
 FLOAT_OP(cvtps, pw)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    FST2 = int32_to_float32(WT0, &env->fp_status);
-    FSTH2 = int32_to_float32(WTH0, &env->fp_status);
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvtps_pw);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvtpw, ps)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    WT2 = float32_to_int32(FST0, &env->fp_status);
-    WTH2 = float32_to_int32(FSTH0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
+    CALL_FROM_TB0(do_float_cvtpw_ps);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvts, d)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    FST2 = float64_to_float32(FDT0, &env->fp_status);
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvts_d);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvts, w)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    FST2 = int32_to_float32(WT0, &env->fp_status);
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvts_w);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvts, l)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    FST2 = int64_to_float32(DT0, &env->fp_status);
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvts_l);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvts, pl)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    WT2 = WT0;
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvts_pl);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvts, pu)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    WT2 = WTH0;
-    update_fcr31();
+    CALL_FROM_TB0(do_float_cvts_pu);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvtw, s)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    WT2 = float32_to_int32(FST0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
+    CALL_FROM_TB0(do_float_cvtw_s);
     DEBUG_FPU_STATE();
     RETURN();
 }
 FLOAT_OP(cvtw, d)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    WT2 = float64_to_int32(FDT0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
+    CALL_FROM_TB0(do_float_cvtw_d);
     DEBUG_FPU_STATE();
     RETURN();
 }
@@ -1917,177 +1824,34 @@ FLOAT_OP(puu, ps)
     RETURN();
 }
 
-FLOAT_OP(roundl, d)
-{
-    set_float_rounding_mode(float_round_nearest_even, &env->fp_status);
-    DT2 = float64_round_to_int(FDT0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(roundl, s)
-{
-    set_float_rounding_mode(float_round_nearest_even, &env->fp_status);
-    DT2 = float32_round_to_int(FST0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(roundw, d)
-{
-    set_float_rounding_mode(float_round_nearest_even, &env->fp_status);
-    WT2 = float64_round_to_int(FDT0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(roundw, s)
-{
-    set_float_rounding_mode(float_round_nearest_even, &env->fp_status);
-    WT2 = float32_round_to_int(FST0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
-    DEBUG_FPU_STATE();
-    RETURN();
+#define FLOAT_ROUNDOP(op, ttype, stype)                    \
+FLOAT_OP(op ## ttype, stype)                               \
+{                                                          \
+    CALL_FROM_TB0(do_float_ ## op ## ttype ## _ ## stype); \
+    DEBUG_FPU_STATE();                                     \
+    RETURN();                                              \
 }
 
-FLOAT_OP(truncl, d)
-{
-    DT2 = float64_to_int64_round_to_zero(FDT0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(truncl, s)
-{
-    DT2 = float32_to_int64_round_to_zero(FST0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(truncw, d)
-{
-    WT2 = float64_to_int32_round_to_zero(FDT0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(truncw, s)
-{
-    WT2 = float32_to_int32_round_to_zero(FST0, &env->fp_status);
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
+FLOAT_ROUNDOP(round, l, d)
+FLOAT_ROUNDOP(round, l, s)
+FLOAT_ROUNDOP(round, w, d)
+FLOAT_ROUNDOP(round, w, s)
 
-FLOAT_OP(ceill, d)
-{
-    set_float_rounding_mode(float_round_up, &env->fp_status);
-    DT2 = float64_round_to_int(FDT0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(ceill, s)
-{
-    set_float_rounding_mode(float_round_up, &env->fp_status);
-    DT2 = float32_round_to_int(FST0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(ceilw, d)
-{
-    set_float_rounding_mode(float_round_up, &env->fp_status);
-    WT2 = float64_round_to_int(FDT0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(ceilw, s)
-{
-    set_float_rounding_mode(float_round_up, &env->fp_status);
-    WT2 = float32_round_to_int(FST0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
+FLOAT_ROUNDOP(trunc, l, d)
+FLOAT_ROUNDOP(trunc, l, s)
+FLOAT_ROUNDOP(trunc, w, d)
+FLOAT_ROUNDOP(trunc, w, s)
 
-FLOAT_OP(floorl, d)
-{
-    set_float_rounding_mode(float_round_down, &env->fp_status);
-    DT2 = float64_round_to_int(FDT0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(floorl, s)
-{
-    set_float_rounding_mode(float_round_down, &env->fp_status);
-    DT2 = float32_round_to_int(FST0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        DT2 = 0x7fffffffffffffffULL;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(floorw, d)
-{
-    set_float_rounding_mode(float_round_down, &env->fp_status);
-    WT2 = float64_round_to_int(FDT0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
-FLOAT_OP(floorw, s)
-{
-    set_float_rounding_mode(float_round_down, &env->fp_status);
-    WT2 = float32_round_to_int(FST0, &env->fp_status);
-    RESTORE_ROUNDING_MODE;
-    update_fcr31();
-    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
-        WT2 = 0x7fffffff;
-    DEBUG_FPU_STATE();
-    RETURN();
-}
+FLOAT_ROUNDOP(ceil, l, d)
+FLOAT_ROUNDOP(ceil, l, s)
+FLOAT_ROUNDOP(ceil, w, d)
+FLOAT_ROUNDOP(ceil, w, s)
+
+FLOAT_ROUNDOP(floor, l, d)
+FLOAT_ROUNDOP(floor, l, s)
+FLOAT_ROUNDOP(floor, w, d)
+FLOAT_ROUNDOP(floor, w, s)
+#undef FLOAR_ROUNDOP
 
 FLOAT_OP(movf, d)
 {
@@ -2186,26 +1950,19 @@ FLOAT_OP(movn, ps)
 #define FLOAT_BINOP(name) \
 FLOAT_OP(name, d)         \
 {                         \
-    set_float_exception_flags(0, &env->fp_status);            \
-    FDT2 = float64_ ## name (FDT0, FDT1, &env->fp_status);    \
-    update_fcr31();       \
+    CALL_FROM_TB0(do_float_ ## name ## _d);  \
     DEBUG_FPU_STATE();    \
     RETURN();             \
 }                         \
 FLOAT_OP(name, s)         \
 {                         \
-    set_float_exception_flags(0, &env->fp_status);            \
-    FST2 = float32_ ## name (FST0, FST1, &env->fp_status);    \
-    update_fcr31();       \
+    CALL_FROM_TB0(do_float_ ## name ## _s);  \
     DEBUG_FPU_STATE();    \
     RETURN();             \
 }                         \
 FLOAT_OP(name, ps)        \
 {                         \
-    set_float_exception_flags(0, &env->fp_status);            \
-    FST2 = float32_ ## name (FST0, FST1, &env->fp_status);    \
-    FSTH2 = float32_ ## name (FSTH0, FSTH1, &env->fp_status); \
-    update_fcr31();       \
+    CALL_FROM_TB0(do_float_ ## name ## _ps); \
     DEBUG_FPU_STATE();    \
     RETURN();             \
 }
@@ -2217,10 +1974,7 @@ FLOAT_BINOP(div)
 
 FLOAT_OP(addr, ps)
 {
-    set_float_exception_flags(0, &env->fp_status);
-    FST2 = float32_add (FST0, FSTH0, &env->fp_status);
-    FSTH2 = float32_add (FST1, FSTH1, &env->fp_status);
-    update_fcr31();
+    CALL_FROM_TB0(do_float_addr_ps);
     DEBUG_FPU_STATE();
     RETURN();
 }
@@ -2390,249 +2144,77 @@ FLOAT_OP(alnv, ps)
 
 extern void dump_fpu_s(CPUState *env);
 
-#define FOP_COND_D(op, cond)                   \
-void op_cmp_d_ ## op (void)                    \
-{                                              \
-    int c = cond;                              \
-    update_fcr31();                            \
-    if (c)                                     \
-        SET_FP_COND(PARAM1, env);              \
-    else                                       \
-        CLEAR_FP_COND(PARAM1, env);            \
-    DEBUG_FPU_STATE();                         \
-    RETURN();                                  \
-}                                              \
-void op_cmpabs_d_ ## op (void)                 \
-{                                              \
-    int c;                                     \
-    FDT0 &= ~(1ULL << 63);                     \
-    FDT1 &= ~(1ULL << 63);                     \
-    c = cond;                                  \
-    update_fcr31();                            \
-    if (c)                                     \
-        SET_FP_COND(PARAM1, env);              \
-    else                                       \
-        CLEAR_FP_COND(PARAM1, env);            \
-    DEBUG_FPU_STATE();                         \
-    RETURN();                                  \
+#define CMP_OP(fmt, op)                                \
+void OPPROTO op_cmp ## _ ## fmt ## _ ## op(void)       \
+{                                                      \
+    CALL_FROM_TB1(do_cmp ## _ ## fmt ## _ ## op, PARAM1); \
+    DEBUG_FPU_STATE();                                 \
+    RETURN();                                          \
+}                                                      \
+void OPPROTO op_cmpabs ## _ ## fmt ## _ ## op(void)    \
+{                                                      \
+    CALL_FROM_TB1(do_cmpabs ## _ ## fmt ## _ ## op, PARAM1); \
+    DEBUG_FPU_STATE();                                 \
+    RETURN();                                          \
 }
+#define CMP_OPS(op)   \
+CMP_OP(d, op)         \
+CMP_OP(s, op)         \
+CMP_OP(ps, op)
 
-int float64_is_unordered(int sig, float64 a, float64 b STATUS_PARAM)
-{
-    if (float64_is_signaling_nan(a) ||
-        float64_is_signaling_nan(b) ||
-        (sig && (float64_is_nan(a) || float64_is_nan(b)))) {
-        float_raise(float_flag_invalid, status);
-        return 1;
-    } else if (float64_is_nan(a) || float64_is_nan(b)) {
-        return 1;
-    } else {
-        return 0;
-    }
-}
-
-/* NOTE: the comma operator will make "cond" to eval to false,
- * but float*_is_unordered() is still called. */
-FOP_COND_D(f,   (float64_is_unordered(0, FDT1, FDT0, &env->fp_status), 0))
-FOP_COND_D(un,  float64_is_unordered(0, FDT1, FDT0, &env->fp_status))
-FOP_COND_D(eq,  !float64_is_unordered(0, FDT1, FDT0, &env->fp_status) && float64_eq(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(ueq, float64_is_unordered(0, FDT1, FDT0, &env->fp_status)  || float64_eq(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(olt, !float64_is_unordered(0, FDT1, FDT0, &env->fp_status) && float64_lt(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(ult, float64_is_unordered(0, FDT1, FDT0, &env->fp_status)  || float64_lt(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(ole, !float64_is_unordered(0, FDT1, FDT0, &env->fp_status) && float64_le(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(ule, float64_is_unordered(0, FDT1, FDT0, &env->fp_status)  || float64_le(FDT0, FDT1, &env->fp_status))
-/* NOTE: the comma operator will make "cond" to eval to false,
- * but float*_is_unordered() is still called. */
-FOP_COND_D(sf,  (float64_is_unordered(1, FDT1, FDT0, &env->fp_status), 0))
-FOP_COND_D(ngle,float64_is_unordered(1, FDT1, FDT0, &env->fp_status))
-FOP_COND_D(seq, !float64_is_unordered(1, FDT1, FDT0, &env->fp_status) && float64_eq(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(ngl, float64_is_unordered(1, FDT1, FDT0, &env->fp_status)  || float64_eq(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(lt,  !float64_is_unordered(1, FDT1, FDT0, &env->fp_status) && float64_lt(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(nge, float64_is_unordered(1, FDT1, FDT0, &env->fp_status)  || float64_lt(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(le,  !float64_is_unordered(1, FDT1, FDT0, &env->fp_status) && float64_le(FDT0, FDT1, &env->fp_status))
-FOP_COND_D(ngt, float64_is_unordered(1, FDT1, FDT0, &env->fp_status)  || float64_le(FDT0, FDT1, &env->fp_status))
-
-#define FOP_COND_S(op, cond)                   \
-void op_cmp_s_ ## op (void)                    \
-{                                              \
-    int c = cond;                              \
-    update_fcr31();                            \
-    if (c)                                     \
-        SET_FP_COND(PARAM1, env);              \
-    else                                       \
-        CLEAR_FP_COND(PARAM1, env);            \
-    DEBUG_FPU_STATE();                         \
-    RETURN();                                  \
-}                                              \
-void op_cmpabs_s_ ## op (void)                 \
-{                                              \
-    int c;                                     \
-    FST0 &= ~(1 << 31);                        \
-    FST1 &= ~(1 << 31);                        \
-    c = cond;                                  \
-    update_fcr31();                            \
-    if (c)                                     \
-        SET_FP_COND(PARAM1, env);              \
-    else                                       \
-        CLEAR_FP_COND(PARAM1, env);            \
-    DEBUG_FPU_STATE();                         \
-    RETURN();                                  \
-}
-
-flag float32_is_unordered(int sig, float32 a, float32 b STATUS_PARAM)
-{
-    extern flag float32_is_nan(float32 a);
-    if (float32_is_signaling_nan(a) ||
-        float32_is_signaling_nan(b) ||
-        (sig && (float32_is_nan(a) || float32_is_nan(b)))) {
-        float_raise(float_flag_invalid, status);
-        return 1;
-    } else if (float32_is_nan(a) || float32_is_nan(b)) {
-        return 1;
-    } else {
-        return 0;
-    }
-}
-
-/* NOTE: the comma operator will make "cond" to eval to false,
- * but float*_is_unordered() is still called. */
-FOP_COND_S(f,   (float32_is_unordered(0, FST1, FST0, &env->fp_status), 0))
-FOP_COND_S(un,  float32_is_unordered(0, FST1, FST0, &env->fp_status))
-FOP_COND_S(eq,  !float32_is_unordered(0, FST1, FST0, &env->fp_status) && float32_eq(FST0, FST1, &env->fp_status))
-FOP_COND_S(ueq, float32_is_unordered(0, FST1, FST0, &env->fp_status)  || float32_eq(FST0, FST1, &env->fp_status))
-FOP_COND_S(olt, !float32_is_unordered(0, FST1, FST0, &env->fp_status) && float32_lt(FST0, FST1, &env->fp_status))
-FOP_COND_S(ult, float32_is_unordered(0, FST1, FST0, &env->fp_status)  || float32_lt(FST0, FST1, &env->fp_status))
-FOP_COND_S(ole, !float32_is_unordered(0, FST1, FST0, &env->fp_status) && float32_le(FST0, FST1, &env->fp_status))
-FOP_COND_S(ule, float32_is_unordered(0, FST1, FST0, &env->fp_status)  || float32_le(FST0, FST1, &env->fp_status))
-/* NOTE: the comma operator will make "cond" to eval to false,
- * but float*_is_unordered() is still called. */
-FOP_COND_S(sf,  (float32_is_unordered(1, FST1, FST0, &env->fp_status), 0))
-FOP_COND_S(ngle,float32_is_unordered(1, FST1, FST0, &env->fp_status))
-FOP_COND_S(seq, !float32_is_unordered(1, FST1, FST0, &env->fp_status) && float32_eq(FST0, FST1, &env->fp_status))
-FOP_COND_S(ngl, float32_is_unordered(1, FST1, FST0, &env->fp_status)  || float32_eq(FST0, FST1, &env->fp_status))
-FOP_COND_S(lt,  !float32_is_unordered(1, FST1, FST0, &env->fp_status) && float32_lt(FST0, FST1, &env->fp_status))
-FOP_COND_S(nge, float32_is_unordered(1, FST1, FST0, &env->fp_status)  || float32_lt(FST0, FST1, &env->fp_status))
-FOP_COND_S(le,  !float32_is_unordered(1, FST1, FST0, &env->fp_status) && float32_le(FST0, FST1, &env->fp_status))
-FOP_COND_S(ngt, float32_is_unordered(1, FST1, FST0, &env->fp_status)  || float32_le(FST0, FST1, &env->fp_status))
-
-#define FOP_COND_PS(op, condl, condh)          \
-void op_cmp_ps_ ## op (void)                   \
-{                                              \
-    int cl = condl;                            \
-    int ch = condh;                            \
-    update_fcr31();                            \
-    if (cl)                                    \
-        SET_FP_COND(PARAM1, env);              \
-    else                                       \
-        CLEAR_FP_COND(PARAM1, env);            \
-    if (ch)                                    \
-        SET_FP_COND(PARAM1 + 1, env);          \
-    else                                       \
-        CLEAR_FP_COND(PARAM1 + 1, env);        \
-    DEBUG_FPU_STATE();                         \
-    RETURN();                                  \
-}                                              \
-void op_cmpabs_ps_ ## op (void)                \
-{                                              \
-    int cl, ch;                                \
-    FST0 &= ~(1 << 31);                        \
-    FSTH0 &= ~(1 << 31);                       \
-    FST1 &= ~(1 << 31);                        \
-    FSTH1 &= ~(1 << 31);                       \
-    cl = condl;                                \
-    ch = condh;                                \
-    update_fcr31();                            \
-    if (cl)                                    \
-        SET_FP_COND(PARAM1, env);              \
-    else                                       \
-        CLEAR_FP_COND(PARAM1, env);            \
-    if (ch)                                    \
-        SET_FP_COND(PARAM1 + 1, env);          \
-    else                                       \
-        CLEAR_FP_COND(PARAM1 + 1, env);        \
-    DEBUG_FPU_STATE();                         \
-    RETURN();                                  \
-}
-
-/* NOTE: the comma operator will make "cond" to eval to false,
- * but float*_is_unordered() is still called. */
-FOP_COND_PS(f,   (float32_is_unordered(0, FST1, FST0, &env->fp_status), 0),
-                 (float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status), 0))
-FOP_COND_PS(un,  float32_is_unordered(0, FST1, FST0, &env->fp_status),
-                 float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status))
-FOP_COND_PS(eq,  !float32_is_unordered(0, FST1, FST0, &env->fp_status)   && float32_eq(FST0, FST1, &env->fp_status),
-                 !float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status) && float32_eq(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(ueq, float32_is_unordered(0, FST1, FST0, &env->fp_status)    || float32_eq(FST0, FST1, &env->fp_status),
-                 float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status)  || float32_eq(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(olt, !float32_is_unordered(0, FST1, FST0, &env->fp_status)   && float32_lt(FST0, FST1, &env->fp_status),
-                 !float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status) && float32_lt(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(ult, float32_is_unordered(0, FST1, FST0, &env->fp_status)    || float32_lt(FST0, FST1, &env->fp_status),
-                 float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status)  || float32_lt(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(ole, !float32_is_unordered(0, FST1, FST0, &env->fp_status)   && float32_le(FST0, FST1, &env->fp_status),
-                 !float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status) && float32_le(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(ule, float32_is_unordered(0, FST1, FST0, &env->fp_status)    || float32_le(FST0, FST1, &env->fp_status),
-                 float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status)  || float32_le(FSTH0, FSTH1, &env->fp_status))
-/* NOTE: the comma operator will make "cond" to eval to false,
- * but float*_is_unordered() is still called. */
-FOP_COND_PS(sf,  (float32_is_unordered(1, FST1, FST0, &env->fp_status), 0),
-                 (float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status), 0))
-FOP_COND_PS(ngle,float32_is_unordered(1, FST1, FST0, &env->fp_status),
-                 float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status))
-FOP_COND_PS(seq, !float32_is_unordered(1, FST1, FST0, &env->fp_status)   && float32_eq(FST0, FST1, &env->fp_status),
-                 !float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status) && float32_eq(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(ngl, float32_is_unordered(1, FST1, FST0, &env->fp_status)    || float32_eq(FST0, FST1, &env->fp_status),
-                 float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status)  || float32_eq(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(lt,  !float32_is_unordered(1, FST1, FST0, &env->fp_status)   && float32_lt(FST0, FST1, &env->fp_status),
-                 !float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status) && float32_lt(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(nge, float32_is_unordered(1, FST1, FST0, &env->fp_status)    || float32_lt(FST0, FST1, &env->fp_status),
-                 float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status)  || float32_lt(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(le,  !float32_is_unordered(1, FST1, FST0, &env->fp_status)   && float32_le(FST0, FST1, &env->fp_status),
-                 !float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status) && float32_le(FSTH0, FSTH1, &env->fp_status))
-FOP_COND_PS(ngt, float32_is_unordered(1, FST1, FST0, &env->fp_status)    || float32_le(FST0, FST1, &env->fp_status),
-                 float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status)  || float32_le(FSTH0, FSTH1, &env->fp_status))
+CMP_OPS(f)
+CMP_OPS(un)
+CMP_OPS(eq)
+CMP_OPS(ueq)
+CMP_OPS(olt)
+CMP_OPS(ult)
+CMP_OPS(ole)
+CMP_OPS(ule)
+CMP_OPS(sf)
+CMP_OPS(ngle)
+CMP_OPS(seq)
+CMP_OPS(ngl)
+CMP_OPS(lt)
+CMP_OPS(nge)
+CMP_OPS(le)
+CMP_OPS(ngt)
+#undef CMP_OPS
+#undef CMP_OP
 
 void op_bc1f (void)
 {
-    T0 = !IS_FP_COND_SET(PARAM1, env);
+    T0 = !!(~GET_FP_COND(env) & (0x1 << PARAM1));
     DEBUG_FPU_STATE();
     RETURN();
 }
-void op_bc1fany2 (void)
+void op_bc1any2f (void)
 {
-    T0 = (!IS_FP_COND_SET(PARAM1, env) ||
-          !IS_FP_COND_SET(PARAM1 + 1, env));
+    T0 = !!(~GET_FP_COND(env) & (0x3 << PARAM1));
     DEBUG_FPU_STATE();
     RETURN();
 }
-void op_bc1fany4 (void)
+void op_bc1any4f (void)
 {
-    T0 = (!IS_FP_COND_SET(PARAM1, env) ||
-          !IS_FP_COND_SET(PARAM1 + 1, env) ||
-          !IS_FP_COND_SET(PARAM1 + 2, env) ||
-          !IS_FP_COND_SET(PARAM1 + 3, env));
+    T0 = !!(~GET_FP_COND(env) & (0xf << PARAM1));
     DEBUG_FPU_STATE();
     RETURN();
 }
 
 void op_bc1t (void)
 {
-    T0 = IS_FP_COND_SET(PARAM1, env);
+    T0 = !!(GET_FP_COND(env) & (0x1 << PARAM1));
     DEBUG_FPU_STATE();
     RETURN();
 }
-void op_bc1tany2 (void)
+void op_bc1any2t (void)
 {
-    T0 = (IS_FP_COND_SET(PARAM1, env) ||
-          IS_FP_COND_SET(PARAM1 + 1, env));
+    T0 = !!(GET_FP_COND(env) & (0x3 << PARAM1));
     DEBUG_FPU_STATE();
     RETURN();
 }
-void op_bc1tany4 (void)
+void op_bc1any4t (void)
 {
-    T0 = (IS_FP_COND_SET(PARAM1, env) ||
-          IS_FP_COND_SET(PARAM1 + 1, env) ||
-          IS_FP_COND_SET(PARAM1 + 2, env) ||
-          IS_FP_COND_SET(PARAM1 + 3, env));
+    T0 = !!(GET_FP_COND(env) & (0xf << PARAM1));
     DEBUG_FPU_STATE();
     RETURN();
 }
@@ -2808,17 +2390,6 @@ void op_save_pc (void)
     RETURN();
 }
 
-void op_save_fp_status (void)
-{
-    union fps {
-        uint32_t i;
-        float_status f;
-    } fps;
-    fps.i = PARAM1;
-    env->fp_status = fps.f;
-    RETURN();
-}
-
 void op_interrupt_restart (void)
 {
     if (!(env->CP0_Status & (1 << CP0St_EXL)) &&
diff --git a/target-mips/op_helper.c b/target-mips/op_helper.c
index 9cc55dcef8..85ad33355a 100644
--- a/target-mips/op_helper.c
+++ b/target-mips/op_helper.c
@@ -598,3 +598,544 @@ void tlb_fill (target_ulong addr, int is_write, int is_user, void *retaddr)
 }
 
 #endif
+
+/* Complex FPU operations which may need stack space. */
+
+/* convert MIPS rounding mode in FCR31 to IEEE library */
+unsigned int ieee_rm[] = {
+    float_round_nearest_even,
+    float_round_to_zero,
+    float_round_up,
+    float_round_down
+};
+
+#define RESTORE_ROUNDING_MODE \
+    set_float_rounding_mode(ieee_rm[env->fcr31 & 3], &env->fp_status)
+
+void do_ctc1 (void)
+{
+    switch(T1) {
+    case 25:
+        if (T0 & 0xffffff00)
+            return;
+        env->fcr31 = (env->fcr31 & 0x017fffff) | ((T0 & 0xfe) << 24) |
+                     ((T0 & 0x1) << 23);
+        break;
+    case 26:
+        if (T0 & 0x007c0000)
+            return;
+        env->fcr31 = (env->fcr31 & 0xfffc0f83) | (T0 & 0x0003f07c);
+        break;
+    case 28:
+        if (T0 & 0x007c0000)
+            return;
+        env->fcr31 = (env->fcr31 & 0xfefff07c) | (T0 & 0x00000f83) |
+                     ((T0 & 0x4) << 22);
+        break;
+    case 31:
+        if (T0 & 0x007c0000)
+            return;
+        env->fcr31 = T0;
+        break;
+    default:
+        return;
+    }
+    /* set rounding mode */
+    RESTORE_ROUNDING_MODE;
+    set_float_exception_flags(0, &env->fp_status);
+    if ((GET_FP_ENABLE(env->fcr31) | 0x20) & GET_FP_CAUSE(env->fcr31))
+        do_raise_exception(EXCP_FPE);
+}
+
+inline char ieee_ex_to_mips(char xcpt)
+{
+    return (xcpt & float_flag_inexact) >> 5 |
+           (xcpt & float_flag_underflow) >> 3 |
+           (xcpt & float_flag_overflow) >> 1 |
+           (xcpt & float_flag_divbyzero) << 1 |
+           (xcpt & float_flag_invalid) << 4;
+}
+
+inline char mips_ex_to_ieee(char xcpt)
+{
+    return (xcpt & FP_INEXACT) << 5 |
+           (xcpt & FP_UNDERFLOW) << 3 |
+           (xcpt & FP_OVERFLOW) << 1 |
+           (xcpt & FP_DIV0) >> 1 |
+           (xcpt & FP_INVALID) >> 4;
+}
+
+inline void update_fcr31(void)
+{
+    int tmp = ieee_ex_to_mips(get_float_exception_flags(&env->fp_status));
+
+    SET_FP_CAUSE(env->fcr31, tmp);
+    if (GET_FP_ENABLE(env->fcr31) & tmp)
+        do_raise_exception(EXCP_FPE);
+    else
+        UPDATE_FP_FLAGS(env->fcr31, tmp);
+}
+
+#define FLOAT_OP(name, p) void do_float_##name##_##p(void)
+
+FLOAT_OP(cvtd, s)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    FDT2 = float32_to_float64(FST0, &env->fp_status);
+    update_fcr31();
+}
+FLOAT_OP(cvtd, w)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    FDT2 = int32_to_float64(WT0, &env->fp_status);
+    update_fcr31();
+}
+FLOAT_OP(cvtd, l)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    FDT2 = int64_to_float64(DT0, &env->fp_status);
+    update_fcr31();
+}
+FLOAT_OP(cvtl, d)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    DT2 = float64_to_int64(FDT0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(cvtl, s)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    DT2 = float32_to_int64(FST0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+
+FLOAT_OP(cvtps, pw)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    FST2 = int32_to_float32(WT0, &env->fp_status);
+    FSTH2 = int32_to_float32(WTH0, &env->fp_status);
+    update_fcr31();
+}
+FLOAT_OP(cvtpw, ps)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    WT2 = float32_to_int32(FST0, &env->fp_status);
+    WTH2 = float32_to_int32(FSTH0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+FLOAT_OP(cvts, d)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    FST2 = float64_to_float32(FDT0, &env->fp_status);
+    update_fcr31();
+}
+FLOAT_OP(cvts, w)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    FST2 = int32_to_float32(WT0, &env->fp_status);
+    update_fcr31();
+}
+FLOAT_OP(cvts, l)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    FST2 = int64_to_float32(DT0, &env->fp_status);
+    update_fcr31();
+}
+FLOAT_OP(cvts, pl)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    WT2 = WT0;
+    update_fcr31();
+}
+FLOAT_OP(cvts, pu)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    WT2 = WTH0;
+    update_fcr31();
+}
+FLOAT_OP(cvtw, s)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    WT2 = float32_to_int32(FST0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+FLOAT_OP(cvtw, d)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    WT2 = float64_to_int32(FDT0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+
+FLOAT_OP(roundl, d)
+{
+    set_float_rounding_mode(float_round_nearest_even, &env->fp_status);
+    DT2 = float64_round_to_int(FDT0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(roundl, s)
+{
+    set_float_rounding_mode(float_round_nearest_even, &env->fp_status);
+    DT2 = float32_round_to_int(FST0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(roundw, d)
+{
+    set_float_rounding_mode(float_round_nearest_even, &env->fp_status);
+    WT2 = float64_round_to_int(FDT0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+FLOAT_OP(roundw, s)
+{
+    set_float_rounding_mode(float_round_nearest_even, &env->fp_status);
+    WT2 = float32_round_to_int(FST0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+
+FLOAT_OP(truncl, d)
+{
+    DT2 = float64_to_int64_round_to_zero(FDT0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(truncl, s)
+{
+    DT2 = float32_to_int64_round_to_zero(FST0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(truncw, d)
+{
+    WT2 = float64_to_int32_round_to_zero(FDT0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+FLOAT_OP(truncw, s)
+{
+    WT2 = float32_to_int32_round_to_zero(FST0, &env->fp_status);
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+
+FLOAT_OP(ceill, d)
+{
+    set_float_rounding_mode(float_round_up, &env->fp_status);
+    DT2 = float64_round_to_int(FDT0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(ceill, s)
+{
+    set_float_rounding_mode(float_round_up, &env->fp_status);
+    DT2 = float32_round_to_int(FST0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(ceilw, d)
+{
+    set_float_rounding_mode(float_round_up, &env->fp_status);
+    WT2 = float64_round_to_int(FDT0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+FLOAT_OP(ceilw, s)
+{
+    set_float_rounding_mode(float_round_up, &env->fp_status);
+    WT2 = float32_round_to_int(FST0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+
+FLOAT_OP(floorl, d)
+{
+    set_float_rounding_mode(float_round_down, &env->fp_status);
+    DT2 = float64_round_to_int(FDT0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(floorl, s)
+{
+    set_float_rounding_mode(float_round_down, &env->fp_status);
+    DT2 = float32_round_to_int(FST0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        DT2 = 0x7fffffffffffffffULL;
+}
+FLOAT_OP(floorw, d)
+{
+    set_float_rounding_mode(float_round_down, &env->fp_status);
+    WT2 = float64_round_to_int(FDT0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+FLOAT_OP(floorw, s)
+{
+    set_float_rounding_mode(float_round_down, &env->fp_status);
+    WT2 = float32_round_to_int(FST0, &env->fp_status);
+    RESTORE_ROUNDING_MODE;
+    update_fcr31();
+    if (GET_FP_CAUSE(env->fcr31) & (FP_OVERFLOW | FP_INVALID))
+        WT2 = 0x7fffffff;
+}
+
+/* binary operations */
+#define FLOAT_BINOP(name) \
+FLOAT_OP(name, d)         \
+{                         \
+    set_float_exception_flags(0, &env->fp_status);            \
+    FDT2 = float64_ ## name (FDT0, FDT1, &env->fp_status);    \
+    update_fcr31();       \
+}                         \
+FLOAT_OP(name, s)         \
+{                         \
+    set_float_exception_flags(0, &env->fp_status);            \
+    FST2 = float32_ ## name (FST0, FST1, &env->fp_status);    \
+    update_fcr31();       \
+}                         \
+FLOAT_OP(name, ps)        \
+{                         \
+    set_float_exception_flags(0, &env->fp_status);            \
+    FST2 = float32_ ## name (FST0, FST1, &env->fp_status);    \
+    FSTH2 = float32_ ## name (FSTH0, FSTH1, &env->fp_status); \
+    update_fcr31();       \
+}
+FLOAT_BINOP(add)
+FLOAT_BINOP(sub)
+FLOAT_BINOP(mul)
+FLOAT_BINOP(div)
+#undef FLOAT_BINOP
+
+FLOAT_OP(addr, ps)
+{
+    set_float_exception_flags(0, &env->fp_status);
+    FST2 = float32_add (FST0, FSTH0, &env->fp_status);
+    FSTH2 = float32_add (FST1, FSTH1, &env->fp_status);
+    update_fcr31();
+}
+
+#define FOP_COND_D(op, cond)                   \
+void do_cmp_d_ ## op (long cc)                 \
+{                                              \
+    int c = cond;                              \
+    update_fcr31();                            \
+    if (c)                                     \
+        SET_FP_COND(cc, env);                  \
+    else                                       \
+        CLEAR_FP_COND(cc, env);                \
+}                                              \
+void do_cmpabs_d_ ## op (long cc)              \
+{                                              \
+    int c;                                     \
+    FDT0 &= ~(1ULL << 63);                     \
+    FDT1 &= ~(1ULL << 63);                     \
+    c = cond;                                  \
+    update_fcr31();                            \
+    if (c)                                     \
+        SET_FP_COND(cc, env);                  \
+    else                                       \
+        CLEAR_FP_COND(cc, env);                \
+}
+
+int float64_is_unordered(int sig, float64 a, float64 b STATUS_PARAM)
+{
+    if (float64_is_signaling_nan(a) ||
+        float64_is_signaling_nan(b) ||
+        (sig && (float64_is_nan(a) || float64_is_nan(b)))) {
+        float_raise(float_flag_invalid, status);
+        return 1;
+    } else if (float64_is_nan(a) || float64_is_nan(b)) {
+        return 1;
+    } else {
+        return 0;
+    }
+}
+
+/* NOTE: the comma operator will make "cond" to eval to false,
+ * but float*_is_unordered() is still called. */
+FOP_COND_D(f,   (float64_is_unordered(0, FDT1, FDT0, &env->fp_status), 0))
+FOP_COND_D(un,  float64_is_unordered(0, FDT1, FDT0, &env->fp_status))
+FOP_COND_D(eq,  !float64_is_unordered(0, FDT1, FDT0, &env->fp_status) && float64_eq(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(ueq, float64_is_unordered(0, FDT1, FDT0, &env->fp_status)  || float64_eq(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(olt, !float64_is_unordered(0, FDT1, FDT0, &env->fp_status) && float64_lt(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(ult, float64_is_unordered(0, FDT1, FDT0, &env->fp_status)  || float64_lt(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(ole, !float64_is_unordered(0, FDT1, FDT0, &env->fp_status) && float64_le(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(ule, float64_is_unordered(0, FDT1, FDT0, &env->fp_status)  || float64_le(FDT0, FDT1, &env->fp_status))
+/* NOTE: the comma operator will make "cond" to eval to false,
+ * but float*_is_unordered() is still called. */
+FOP_COND_D(sf,  (float64_is_unordered(1, FDT1, FDT0, &env->fp_status), 0))
+FOP_COND_D(ngle,float64_is_unordered(1, FDT1, FDT0, &env->fp_status))
+FOP_COND_D(seq, !float64_is_unordered(1, FDT1, FDT0, &env->fp_status) && float64_eq(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(ngl, float64_is_unordered(1, FDT1, FDT0, &env->fp_status)  || float64_eq(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(lt,  !float64_is_unordered(1, FDT1, FDT0, &env->fp_status) && float64_lt(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(nge, float64_is_unordered(1, FDT1, FDT0, &env->fp_status)  || float64_lt(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(le,  !float64_is_unordered(1, FDT1, FDT0, &env->fp_status) && float64_le(FDT0, FDT1, &env->fp_status))
+FOP_COND_D(ngt, float64_is_unordered(1, FDT1, FDT0, &env->fp_status)  || float64_le(FDT0, FDT1, &env->fp_status))
+
+#define FOP_COND_S(op, cond)                   \
+void do_cmp_s_ ## op (long cc)                 \
+{                                              \
+    int c = cond;                              \
+    update_fcr31();                            \
+    if (c)                                     \
+        SET_FP_COND(cc, env);                  \
+    else                                       \
+        CLEAR_FP_COND(cc, env);                \
+}                                              \
+void do_cmpabs_s_ ## op (long cc)              \
+{                                              \
+    int c;                                     \
+    FST0 &= ~(1 << 31);                        \
+    FST1 &= ~(1 << 31);                        \
+    c = cond;                                  \
+    update_fcr31();                            \
+    if (c)                                     \
+        SET_FP_COND(cc, env);                  \
+    else                                       \
+        CLEAR_FP_COND(cc, env);                \
+}
+
+flag float32_is_unordered(int sig, float32 a, float32 b STATUS_PARAM)
+{
+    extern flag float32_is_nan(float32 a);
+    if (float32_is_signaling_nan(a) ||
+        float32_is_signaling_nan(b) ||
+        (sig && (float32_is_nan(a) || float32_is_nan(b)))) {
+        float_raise(float_flag_invalid, status);
+        return 1;
+    } else if (float32_is_nan(a) || float32_is_nan(b)) {
+        return 1;
+    } else {
+        return 0;
+    }
+}
+
+/* NOTE: the comma operator will make "cond" to eval to false,
+ * but float*_is_unordered() is still called. */
+FOP_COND_S(f,   (float32_is_unordered(0, FST1, FST0, &env->fp_status), 0))
+FOP_COND_S(un,  float32_is_unordered(0, FST1, FST0, &env->fp_status))
+FOP_COND_S(eq,  !float32_is_unordered(0, FST1, FST0, &env->fp_status) && float32_eq(FST0, FST1, &env->fp_status))
+FOP_COND_S(ueq, float32_is_unordered(0, FST1, FST0, &env->fp_status)  || float32_eq(FST0, FST1, &env->fp_status))
+FOP_COND_S(olt, !float32_is_unordered(0, FST1, FST0, &env->fp_status) && float32_lt(FST0, FST1, &env->fp_status))
+FOP_COND_S(ult, float32_is_unordered(0, FST1, FST0, &env->fp_status)  || float32_lt(FST0, FST1, &env->fp_status))
+FOP_COND_S(ole, !float32_is_unordered(0, FST1, FST0, &env->fp_status) && float32_le(FST0, FST1, &env->fp_status))
+FOP_COND_S(ule, float32_is_unordered(0, FST1, FST0, &env->fp_status)  || float32_le(FST0, FST1, &env->fp_status))
+/* NOTE: the comma operator will make "cond" to eval to false,
+ * but float*_is_unordered() is still called. */
+FOP_COND_S(sf,  (float32_is_unordered(1, FST1, FST0, &env->fp_status), 0))
+FOP_COND_S(ngle,float32_is_unordered(1, FST1, FST0, &env->fp_status))
+FOP_COND_S(seq, !float32_is_unordered(1, FST1, FST0, &env->fp_status) && float32_eq(FST0, FST1, &env->fp_status))
+FOP_COND_S(ngl, float32_is_unordered(1, FST1, FST0, &env->fp_status)  || float32_eq(FST0, FST1, &env->fp_status))
+FOP_COND_S(lt,  !float32_is_unordered(1, FST1, FST0, &env->fp_status) && float32_lt(FST0, FST1, &env->fp_status))
+FOP_COND_S(nge, float32_is_unordered(1, FST1, FST0, &env->fp_status)  || float32_lt(FST0, FST1, &env->fp_status))
+FOP_COND_S(le,  !float32_is_unordered(1, FST1, FST0, &env->fp_status) && float32_le(FST0, FST1, &env->fp_status))
+FOP_COND_S(ngt, float32_is_unordered(1, FST1, FST0, &env->fp_status)  || float32_le(FST0, FST1, &env->fp_status))
+
+#define FOP_COND_PS(op, condl, condh)          \
+void do_cmp_ps_ ## op (long cc)                \
+{                                              \
+    int cl = condl;                            \
+    int ch = condh;                            \
+    update_fcr31();                            \
+    if (cl)                                    \
+        SET_FP_COND(cc, env);                  \
+    else                                       \
+        CLEAR_FP_COND(cc, env);                \
+    if (ch)                                    \
+        SET_FP_COND(cc + 1, env);              \
+    else                                       \
+        CLEAR_FP_COND(cc + 1, env);            \
+}                                              \
+void do_cmpabs_ps_ ## op (long cc)             \
+{                                              \
+    int cl, ch;                                \
+    FST0 &= ~(1 << 31);                        \
+    FSTH0 &= ~(1 << 31);                       \
+    FST1 &= ~(1 << 31);                        \
+    FSTH1 &= ~(1 << 31);                       \
+    cl = condl;                                \
+    ch = condh;                                \
+    update_fcr31();                            \
+    if (cl)                                    \
+        SET_FP_COND(cc, env);                  \
+    else                                       \
+        CLEAR_FP_COND(cc, env);                \
+    if (ch)                                    \
+        SET_FP_COND(cc + 1, env);              \
+    else                                       \
+        CLEAR_FP_COND(cc + 1, env);            \
+}
+
+/* NOTE: the comma operator will make "cond" to eval to false,
+ * but float*_is_unordered() is still called. */
+FOP_COND_PS(f,   (float32_is_unordered(0, FST1, FST0, &env->fp_status), 0),
+                 (float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status), 0))
+FOP_COND_PS(un,  float32_is_unordered(0, FST1, FST0, &env->fp_status),
+                 float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status))
+FOP_COND_PS(eq,  !float32_is_unordered(0, FST1, FST0, &env->fp_status)   && float32_eq(FST0, FST1, &env->fp_status),
+                 !float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status) && float32_eq(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(ueq, float32_is_unordered(0, FST1, FST0, &env->fp_status)    || float32_eq(FST0, FST1, &env->fp_status),
+                 float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status)  || float32_eq(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(olt, !float32_is_unordered(0, FST1, FST0, &env->fp_status)   && float32_lt(FST0, FST1, &env->fp_status),
+                 !float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status) && float32_lt(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(ult, float32_is_unordered(0, FST1, FST0, &env->fp_status)    || float32_lt(FST0, FST1, &env->fp_status),
+                 float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status)  || float32_lt(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(ole, !float32_is_unordered(0, FST1, FST0, &env->fp_status)   && float32_le(FST0, FST1, &env->fp_status),
+                 !float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status) && float32_le(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(ule, float32_is_unordered(0, FST1, FST0, &env->fp_status)    || float32_le(FST0, FST1, &env->fp_status),
+                 float32_is_unordered(0, FSTH1, FSTH0, &env->fp_status)  || float32_le(FSTH0, FSTH1, &env->fp_status))
+/* NOTE: the comma operator will make "cond" to eval to false,
+ * but float*_is_unordered() is still called. */
+FOP_COND_PS(sf,  (float32_is_unordered(1, FST1, FST0, &env->fp_status), 0),
+                 (float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status), 0))
+FOP_COND_PS(ngle,float32_is_unordered(1, FST1, FST0, &env->fp_status),
+                 float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status))
+FOP_COND_PS(seq, !float32_is_unordered(1, FST1, FST0, &env->fp_status)   && float32_eq(FST0, FST1, &env->fp_status),
+                 !float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status) && float32_eq(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(ngl, float32_is_unordered(1, FST1, FST0, &env->fp_status)    || float32_eq(FST0, FST1, &env->fp_status),
+                 float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status)  || float32_eq(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(lt,  !float32_is_unordered(1, FST1, FST0, &env->fp_status)   && float32_lt(FST0, FST1, &env->fp_status),
+                 !float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status) && float32_lt(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(nge, float32_is_unordered(1, FST1, FST0, &env->fp_status)    || float32_lt(FST0, FST1, &env->fp_status),
+                 float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status)  || float32_lt(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(le,  !float32_is_unordered(1, FST1, FST0, &env->fp_status)   && float32_le(FST0, FST1, &env->fp_status),
+                 !float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status) && float32_le(FSTH0, FSTH1, &env->fp_status))
+FOP_COND_PS(ngt, float32_is_unordered(1, FST1, FST0, &env->fp_status)    || float32_le(FST0, FST1, &env->fp_status),
+                 float32_is_unordered(1, FSTH1, FSTH0, &env->fp_status)  || float32_le(FSTH0, FSTH1, &env->fp_status))
diff --git a/target-mips/translate.c b/target-mips/translate.c
index 66e9def854..57e5777ccb 100644
--- a/target-mips/translate.c
+++ b/target-mips/translate.c
@@ -491,7 +491,7 @@ FGEN32(gen_op_load_fpr_WTH2,  gen_op_load_fpr_WTH2_fpr);
 FGEN32(gen_op_store_fpr_WTH2, gen_op_store_fpr_WTH2_fpr);
 
 #define FOP_CONDS(type, fmt)                                            \
-static GenOpFunc1 * cond ## type ## _ ## fmt ## _table[16] = {          \
+static GenOpFunc1 * gen_op_cmp ## type ## _ ## fmt ## _table[16] = {    \
     gen_op_cmp ## type ## _ ## fmt ## _f,                               \
     gen_op_cmp ## type ## _ ## fmt ## _un,                              \
     gen_op_cmp ## type ## _ ## fmt ## _eq,                              \
@@ -511,7 +511,7 @@ static GenOpFunc1 * cond ## type ## _ ## fmt ## _table[16] = {          \
 };                                                                      \
 static inline void gen_cmp ## type ## _ ## fmt(int n, long cc)          \
 {                                                                       \
-    cond ## type ## _ ## fmt ## _table[n](cc);                          \
+    gen_op_cmp ## type ## _ ## fmt ## _table[n](cc);                    \
 }
 
 FOP_CONDS(, d)
@@ -525,11 +525,10 @@ typedef struct DisasContext {
     struct TranslationBlock *tb;
     target_ulong pc, saved_pc;
     uint32_t opcode;
-    uint32_t fp_status, saved_fp_status;
+    uint32_t fp_status;
     /* Routine used to access memory */
     int mem_idx;
     uint32_t hflags, saved_hflags;
-    uint32_t CP0_Status;
     int bstate;
     target_ulong btarget;
 } DisasContext;
@@ -628,11 +627,21 @@ static inline void save_cpu_state (DisasContext *ctx, int do_save_pc)
     }
 }
 
-static inline void save_fpu_state (DisasContext *ctx)
+static inline void restore_cpu_state (CPUState *env, DisasContext *ctx)
 {
-    if (ctx->fp_status != ctx->saved_fp_status) {
-        gen_op_save_fp_status(ctx->fp_status);
-        ctx->saved_fp_status = ctx->fp_status;
+    ctx->saved_hflags = ctx->hflags;
+    switch (ctx->hflags & MIPS_HFLAG_BMASK) {
+    case MIPS_HFLAG_BR:
+        gen_op_restore_breg_target();
+        break;
+    case MIPS_HFLAG_B:
+        ctx->btarget = env->btarget;
+        break;
+    case MIPS_HFLAG_BC:
+    case MIPS_HFLAG_BL:
+        ctx->btarget = env->btarget;
+        gen_op_restore_bcond();
+        break;
     }
 }
 
@@ -4293,20 +4302,20 @@ static void gen_compute_branch1 (DisasContext *ctx, uint32_t op,
         gen_op_save_bcond();
         break;
     case OPC_BC1FANY2:
-        gen_op_bc1fany2(cc);
-        opn = "bc1fany2";
+        gen_op_bc1any2f(cc);
+        opn = "bc1any2f";
         goto not_likely;
     case OPC_BC1TANY2:
-        gen_op_bc1tany2(cc);
-        opn = "bc1tany2";
+        gen_op_bc1any2t(cc);
+        opn = "bc1any2t";
         goto not_likely;
     case OPC_BC1FANY4:
-        gen_op_bc1fany4(cc);
-        opn = "bc1fany4";
+        gen_op_bc1any4f(cc);
+        opn = "bc1any4f";
         goto not_likely;
     case OPC_BC1TANY4:
-        gen_op_bc1tany4(cc);
-        opn = "bc1tany4";
+        gen_op_bc1any4t(cc);
+        opn = "bc1any4t";
     not_likely:
         ctx->hflags |= MIPS_HFLAG_BC;
         gen_op_set_bcond();
@@ -4323,27 +4332,6 @@ static void gen_compute_branch1 (DisasContext *ctx, uint32_t op,
 
 /* Coprocessor 1 (FPU) */
 
-/* verify if floating point register is valid; an operation is not defined
- * if bit 0 of any register specification is set and the FR bit in the
- * Status register equals zero, since the register numbers specify an
- * even-odd pair of adjacent coprocessor general registers. When the FR bit
- * in the Status register equals one, both even and odd register numbers
- * are valid. This limitation exists only for 64 bit wide (d,l,ps) registers.
- * 
- * Multiple 64 bit wide registers can be checked by calling
- * CHECK_FR(ctx, freg1 | freg2 | ... | fregN);
- * 
- * FIXME: This is broken for R2, it needs to be checked at runtime, not
- * at translation time.
- */
-#define CHECK_FR(ctx, freg) do {                                      \
-        if (!((ctx)->CP0_Status & (1 << CP0St_FR)) && ((freg) & 1)) { \
-            MIPS_INVAL("FPU mode");                                   \
-            generate_exception (ctx, EXCP_RI);                        \
-            return;                                                   \
-        }                                                             \
-    } while(0)
-
 #define FOP(func, fmt) (((fmt) << 21) | (func))
 
 static void gen_cp1 (DisasContext *ctx, uint32_t opc, int rt, int fs)
@@ -4388,14 +4376,14 @@ static void gen_cp1 (DisasContext *ctx, uint32_t opc, int rt, int fs)
         opn = "dmtc1";
         break;
     case OPC_MFHC1:
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         gen_op_mfhc1();
         GEN_STORE_TN_REG(rt, T0);
         opn = "mfhc1";
         break;
     case OPC_MTHC1:
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_REG_TN(T0, rt);
         gen_op_mthc1();
         GEN_STORE_FTN_FREG(fs, WTH0);
@@ -4546,28 +4534,28 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "neg.s";
         break;
     case FOP(8, 16):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(WT0, fs);
         gen_op_float_roundl_s();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "round.l.s";
         break;
     case FOP(9, 16):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(WT0, fs);
         gen_op_float_truncl_s();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "trunc.l.s";
         break;
     case FOP(10, 16):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(WT0, fs);
         gen_op_float_ceill_s();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "ceil.l.s";
         break;
     case FOP(11, 16):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(WT0, fs);
         gen_op_float_floorl_s();
         GEN_STORE_FTN_FREG(fd, DT2);
@@ -4622,7 +4610,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "movn.s";
         break;
     case FOP(33, 16):
-        CHECK_FR(ctx, fd);
+        gen_op_cp1_registers(fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         gen_op_float_cvtd_s();
         GEN_STORE_FTN_FREG(fd, DT2);
@@ -4635,14 +4623,14 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "cvt.w.s";
         break;
     case FOP(37, 16):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         gen_op_float_cvtl_s();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "cvt.l.s";
         break;
     case FOP(38, 16):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(WT1, fs);
         GEN_LOAD_FREG_FTN(WT0, ft);
         gen_op_float_cvtps_s();
@@ -4676,7 +4664,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         }
         break;
     case FOP(0, 17):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         GEN_LOAD_FREG_FTN(DT1, ft);
         gen_op_float_add_d();
@@ -4685,7 +4673,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         optype = BINOP;
         break;
     case FOP(1, 17):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         GEN_LOAD_FREG_FTN(DT1, ft);
         gen_op_float_sub_d();
@@ -4694,7 +4682,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         optype = BINOP;
         break;
     case FOP(2, 17):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         GEN_LOAD_FREG_FTN(DT1, ft);
         gen_op_float_mul_d();
@@ -4703,7 +4691,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         optype = BINOP;
         break;
     case FOP(3, 17):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         GEN_LOAD_FREG_FTN(DT1, ft);
         gen_op_float_div_d();
@@ -4712,84 +4700,84 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         optype = BINOP;
         break;
     case FOP(4, 17):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_sqrt_d();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "sqrt.d";
         break;
     case FOP(5, 17):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_abs_d();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "abs.d";
         break;
     case FOP(6, 17):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_mov_d();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "mov.d";
         break;
     case FOP(7, 17):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_chs_d();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "neg.d";
         break;
     case FOP(8, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_roundl_d();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "round.l.d";
         break;
     case FOP(9, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_truncl_d();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "trunc.l.d";
         break;
     case FOP(10, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_ceill_d();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "ceil.l.d";
         break;
     case FOP(11, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_floorl_d();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "floor.l.d";
         break;
     case FOP(12, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_roundw_d();
         GEN_STORE_FTN_FREG(fd, WT2);
         opn = "round.w.d";
         break;
     case FOP(13, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_truncw_d();
         GEN_STORE_FTN_FREG(fd, WT2);
         opn = "trunc.w.d";
         break;
     case FOP(14, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_ceilw_d();
         GEN_STORE_FTN_FREG(fd, WT2);
         opn = "ceil.w.d";
         break;
     case FOP(15, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_floorw_d();
         GEN_STORE_FTN_FREG(fd, WT2);
@@ -4835,7 +4823,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
     case FOP(61, 17):
     case FOP(62, 17):
     case FOP(63, 17):
-        CHECK_FR(ctx, fs | ft);
+        gen_op_cp1_registers(fs | ft);
         GEN_LOAD_FREG_FTN(DT0, fs);
         GEN_LOAD_FREG_FTN(DT1, ft);
         if (ctx->opcode & (1 << 6)) {
@@ -4847,21 +4835,21 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         }
         break;
     case FOP(32, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_cvts_d();
         GEN_STORE_FTN_FREG(fd, WT2);
         opn = "cvt.s.d";
         break;
     case FOP(36, 17):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_cvtw_d();
         GEN_STORE_FTN_FREG(fd, WT2);
         opn = "cvt.w.d";
         break;
     case FOP(37, 17):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_cvtl_d();
         GEN_STORE_FTN_FREG(fd, DT2);
@@ -4874,21 +4862,21 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "cvt.s.w";
         break;
     case FOP(33, 20):
-        CHECK_FR(ctx, fd);
+        gen_op_cp1_registers(fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         gen_op_float_cvtd_w();
         GEN_STORE_FTN_FREG(fd, DT2);
         opn = "cvt.d.w";
         break;
     case FOP(32, 21):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_cvts_l();
         GEN_STORE_FTN_FREG(fd, WT2);
         opn = "cvt.s.l";
         break;
     case FOP(33, 21):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(DT0, fs);
         gen_op_float_cvtd_l();
         GEN_STORE_FTN_FREG(fd, DT2);
@@ -4896,7 +4884,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         break;
     case FOP(38, 20):
     case FOP(38, 21):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         gen_op_float_cvtps_pw();
@@ -4905,7 +4893,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "cvt.ps.pw";
         break;
     case FOP(0, 22):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         GEN_LOAD_FREG_FTN(WT1, ft);
@@ -4916,7 +4904,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "add.ps";
         break;
     case FOP(1, 22):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         GEN_LOAD_FREG_FTN(WT1, ft);
@@ -4927,7 +4915,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "sub.ps";
         break;
     case FOP(2, 22):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         GEN_LOAD_FREG_FTN(WT1, ft);
@@ -4938,7 +4926,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "mul.ps";
         break;
     case FOP(5, 22):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         gen_op_float_abs_ps();
@@ -4947,7 +4935,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "abs.ps";
         break;
     case FOP(6, 22):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         gen_op_float_mov_ps();
@@ -4956,7 +4944,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "mov.ps";
         break;
     case FOP(7, 22):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         gen_op_float_chs_ps();
@@ -4998,7 +4986,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "movn.ps";
         break;
     case FOP(24, 22):
-        CHECK_FR(ctx, fs | fd | ft);
+        gen_op_cp1_registers(fs | fd | ft);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         GEN_LOAD_FREG_FTN(WT1, ft);
@@ -5009,14 +4997,14 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "addr.ps";
         break;
     case FOP(32, 22):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         gen_op_float_cvts_pu();
         GEN_STORE_FTN_FREG(fd, WT2);
         opn = "cvt.s.pu";
         break;
     case FOP(36, 22):
-        CHECK_FR(ctx, fs | fd);
+        gen_op_cp1_registers(fs | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         gen_op_float_cvtpw_ps();
@@ -5025,14 +5013,14 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "cvt.pw.ps";
         break;
     case FOP(40, 22):
-        CHECK_FR(ctx, fs);
+        gen_op_cp1_registers(fs);
         GEN_LOAD_FREG_FTN(WT0, fs);
         gen_op_float_cvts_pl();
         GEN_STORE_FTN_FREG(fd, WT2);
         opn = "cvt.s.pl";
         break;
     case FOP(44, 22):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WT1, ft);
         gen_op_float_pll_ps();
@@ -5040,7 +5028,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "pll.ps";
         break;
     case FOP(45, 22):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH1, ft);
         gen_op_float_plu_ps();
@@ -5048,7 +5036,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "plu.ps";
         break;
     case FOP(46, 22):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         GEN_LOAD_FREG_FTN(WT1, ft);
         gen_op_float_pul_ps();
@@ -5056,7 +5044,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
         opn = "pul.ps";
         break;
     case FOP(47, 22):
-        CHECK_FR(ctx, fs | ft | fd);
+        gen_op_cp1_registers(fs | ft | fd);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         GEN_LOAD_FREG_FTN(WTH1, ft);
         gen_op_float_puu_ps();
@@ -5079,7 +5067,7 @@ static void gen_farith (DisasContext *ctx, uint32_t op1, int ft,
     case FOP(61, 22):
     case FOP(62, 22):
     case FOP(63, 22):
-        CHECK_FR(ctx, fs | ft);
+        gen_op_cp1_registers(fs | ft);
         GEN_LOAD_FREG_FTN(WT0, fs);
         GEN_LOAD_FREG_FTN(WTH0, fs);
         GEN_LOAD_FREG_FTN(WT1, ft);
@@ -5166,7 +5154,7 @@ static void gen_flt3_arith (DisasContext *ctx, uint32_t opc, int fd,
     const char *opn = "flt3_arith";
 
     /* All of those work only on 64bit FPUs. */
-    CHECK_FR(ctx, fd | fr | fs | ft);
+    gen_op_cp1_registers(fd | fr | fs | ft);
     switch (opc) {
     case OPC_ALNV_PS:
         GEN_LOAD_REG_TN(T0, fr);
@@ -5874,26 +5862,12 @@ gen_intermediate_code_internal (CPUState *env, TranslationBlock *tb,
     ctx.bstate = BS_NONE;
     /* Restore delay slot state from the tb context.  */
     ctx.hflags = tb->flags;
-    ctx.saved_hflags = ctx.hflags;
-    switch (ctx.hflags & MIPS_HFLAG_BMASK) {
-    case MIPS_HFLAG_BR:
-        gen_op_restore_breg_target();
-        break;
-    case MIPS_HFLAG_B:
-        ctx.btarget = env->btarget;
-        break;
-    case MIPS_HFLAG_BC:
-    case MIPS_HFLAG_BL:
-        ctx.btarget = env->btarget;
-        gen_op_restore_bcond();
-        break;
-    }
+    restore_cpu_state(env, &ctx);
 #if defined(CONFIG_USER_ONLY)
     ctx.mem_idx = 0;
 #else
     ctx.mem_idx = !((ctx.hflags & MIPS_HFLAG_MODE) == MIPS_HFLAG_UM);
 #endif
-    ctx.CP0_Status = env->CP0_Status;
 #ifdef DEBUG_DISAS
     if (loglevel & CPU_LOG_TB_CPU) {
         fprintf(logfile, "------------------------------------------------\n");